A gender perspective on the global migration of scholars

Significance Within a globalizing scientific system, international migration is increasingly recognized as a strategy for scientists to advance their careers. The migration literature more broadly has suggested a process of feminization, with an increasing share of women among all international migrants. With respect to the migration of scholars, however, whether male and female scholars participate equally in transnational mobility and how these patterns have shifted over time from a global perspective are not known. Using bibliometric data that cover the past two decades, we show that, while female researchers continued to be underrepresented among internationally mobile researchers, and migrated over shorter distances, this gender gap has been narrowing at a faster rate than the gender gap in the population of general researchers.

results for a portion of the countries or regions are shown in Table S2. In general, the precision rate and the recall rate were over 90%, which can be seen as proof of the overall high degree of accuracy of our gender detection method. China was an 66 exception, mainly because in the name-gender dictionary, a large share of Chinese first names were unisex in the transliterated 67 Roman alphabet. 68 For the second database, the precision rate and the recall rate for the female authors' first-names were 0.97 and 0.80, 69 respectively; while the precision rate and recall rate for the male authors' first-names were 0.98 and 0.86, respectively. 70 To compare our gender detection method with methods used by other research, we applied another name-gender database 71 generated by a gender disambiguation algorithm (5) in detecting the genders of the researchers in our dataset. This reference Additionally, given the hypothesis that the migration spreads varied across countries over time, different countries have 81 played different roles in the global migration spread. We also measured the weighted average of the country-specific emigration 82 spread (ES W eighted t , defined in Eq. (4)) to represent the overall level of emigration spread during the period t. Accordingly, the 83 global weighted average of the country-specific immigration spread (IS W eighted t , defined in Eq. (5)) can be calculated.
Gender differences in migration distributions. To a given destination country i, the average gender difference among migration 87 inflows at a specific period of time t is measured using Eq. (6),

Further analysis and results
Migration intensity of mobile researchers by gender. The dataset we processed for our analysis on gender disparities in global   104   scholarly migration involved 180, 191, 197, and 202 countries during the periods 1998-2002, 2003-2007, 2008-2012, and   105 2013-2017, respectively. To make the results more comparable, and to capture the main flows and dynamics of global scholarly 106 migration over time, we estimated the migration intensity of mobile researchers by gender which includes both the absolute 107 number of female (male) mobile researchers and their proportions relative to all female (male) researchers (relative number).
108 Table S3 shows the results of the 18 major countries, which were the origin and also the destination countries with the largest 109 50 migration flows in any of the four periods. And these migration flows across the 18 countries accounted for 94.97%, 93.57%, 110 88.69%, and 80.88% of all scholarly migration flows over the four periods, respectively. The Table also indicates that global 111 scholarly migration mainly occurred across a small fraction of countries, which led to the skweness of migration flows.

112
It is evident that the share of scholars and the size of the science system were much larger in some countries than others 113 (e.g., see the case of the USA, which has hosted around one-third of the share of worldwide scholars in all time periods). The 114 countries with the larger and more established science systems greatly influenced the observed migration trends.

115
Gender disparities in transnational emigration and immigration . The gender ratios varied among the outgoing and the incoming 116 researchers, and thus contributed differently to the country-level gender disparities. Meanwhile, the overall gender ratio among 117 migrants may obscure the divergent outcomes for transnational scholarly migrations in both directions. Fig. S1 shows the 118 gender ratios among the outflows and the inflows separately to help us gain a better understanding of how the gendered 119 patterns of scholarly mobility varied across the four time periods.

120
Generally, the countries appearing in all four figure panels experienced stable and slow growth in the female-to-male ratios 121 in both directions. In addition, the countries emerging in the sub-plots of recent periods (especially period three and period 122 four) that had sufficient numbers of female migrant researchers were more likely to display identifiable patterns of gender ratios 123 far from the median values. These countries either had significant gender disparities (such as Pakistan and Saudi Arabia) or 124 relatively balanced gender relationships (such as Portugal and Serbia). Taken as a whole, the gender patterns of scholarly 125 inflows and outflows across most major countries were consistent over time 126 As outliers in the overall gender patterns of the scholarly inflows and outflows, several countries stood out as having a 127 skewed gender ratio in one direction. One example is Brazil. In the most recent period of 2013-2017, around seven female 128 researchers were leaving Brazil for every 10 male researchers who were emigrating. However, the number of female researchers 129 moving to Brazil was nearly half that of male researchers. This indicates that the level of gender balance was higher among the 130 researchers leaving Brazil than among the researchers moving to Brazil.

131
Migration distance by gender. We measured the migration distances based on the geographic distance between the capital cities 132 of the origin and destination countries. As the Fig. S2 shows, there is a gap between the average distance of male and female distance for male researchers still increased but at a slower rate, while female researchers on average tended to migrate shorter 136 distance compared to the previous period. Accordingly, the gender difference regarding the average migration distance became 137 wider. 138 We also compared the average migration distances among female and male researchers across six selected countries in to believe our findings are not skewed by the composition of those researchers for whom we cannot infer gender by names.